My sorry attempt at a review - DS1104Z

[marmad]

One nice feature that the DS1000Z has - which both the DS2000/4000 are missing - is the ability to manually turn ON/OFF sin(x)/x interpolation. On the 2000/4000, this is always automatic.

I'm curious if this puts the decision on which interpolation to use fully in the hands of the user - or whether the DSO sometimes overrides the setting. As usual, Rigol's documentation is lacking.

[cybermaus]

Yes, I do seem to have full control over sin(x)/x on/off (aquire menu), as well as separately over dot/vector (display menu).

EDIT: correction: I can only control sin(x)/x when displaying 3 or 4 channels. On all timescales, with 3/4 channels I can turn it on/off, with 1/2 channels, I cannot control it and is is grayed out "on"

In dot mode sin(x)/x does not do anything of course. But to be honest I am not so sure sin(x)/x is useful for vector either. I may try to keep it off just because of the notion that it saves on CPU power.

Screenprints: 1, 2, 3, 4
4-channel on (so 250Ms per channel) feeding one of them a 12.5+25MHz input sine

1: Dot mode (either sin(x) on or of, does not matter), single shot
2: Dot mode (either sin(x) on or of, does not matter), run/stop on full frame-rate (so sort of with equivalent sampling)v
3: vector mode, sin(x) on, single shot
4: vector mode, sin(x) off, single shot
5: vector mode, sin(x) ??, run/stop (forgot how it was set)

Bonus: Dot mode with only one channel

[marmad]

In dot mode sin(x)/x does not do anything of course. But to be honest I am not so sure sin(x)/x is useful for vector either. I may try to keep it off just because of the notion that it saves on CPU power.

Assuming that the DS1074/1104Z have a similar input filter to the DS2072/2102, you would probably be best served by not using sin(x)/x when sampling rates <=250MSa/s (unless you use the 20MHz BW limit). But for the 500MSa/s and 1GSa/s rates, the sin(x)/x interpolation will give you more accurate waveform reconstruction with smaller sample sizes.

[Mark_O]

Some confusing commentary here.

As far as the efficacy of those 4 channels go, it really depends on what your work is. The effective reliably-interpolated maximum analog bandwidth of the DSO when using 4 channels is ~25MHz (well, ~31.5MHz if you're charitable)...

Huh?  I don't recall you suggesting the "effective reliably-interpolated maximum analog bandwidth" of the DS2072 (modded to 200MHz) was only 100MHz?  (Whose max sampling rate with 2 channels enabled is 1GHz.)  So why have you chosen to select a 10x oversampling qualifier here?

Definitely handy for certain situations (like 4-channel protocol decoding) - but it's not a 4-channel 100MHz BW DSO.

Again, why?  If running at it's max 250MHz sample rate, that's certainly adequate for 100MHz on all 4 channels.  When sampling slower, of course, it's not. 

250Ms/s is enough for 100MHz but it depends on the interpolation whether you get a pretty picture or not.

I understood what you meant, though marmad did not.  If it does sin(x)/x properly (and perhaps it doesn't??, which would explain marmad's comments), then you get the "pretty picture", aka, accurate waveform display, you referenced.

The more pressing problem is aliasing and I'm pretty sure you'll see lots of aliasing on this scope with 4 channels on.

Why are you sure there will be lots of aliasing?  While it's certainly possible to have the sample rate set too low, while examining a signal source with high-frequency components that will cause aliasing, that's true of any scope.

The sampling rate is almost certainly too low for reliable sin(x)/x - and 2.5 samples per period are not enough for an accurate linear reconstruction, pretty or not.

I agree with the later, but don't understand the former?

[mrflibble]

Thanks for this video!  I found your English easy enough to follow, so wouldn't worry about that.

As a prospective 1074Z buyer I like to know all the good & bad before I really really buy one, and your video highlighted a few things I hadn't seen in other vids yet. GUI sluggishness would be a bit of a concern, because that is one of the many reasons why I'm looking for a new scope. The old one is very nice, but it's got the drawbacks of all the old DSO's: low memory depth, and bleeping slow CPU (68020). Now the 68k was great fun in the Amiga era, but uhm, yeah... it's a bit slow compared to modern stuff.

[marmad]

Huh?  I don't recall you suggesting the "effective reliably-interpolated maximum analog bandwidth" of the DS2072 (modded to 200MHz) was only 100MHz?  (Whose max sampling rate with 2 channels enabled is 1GHz.)  So why have you chosen to select a 10x oversampling qualifier here?

Actually, I've been suggesting exactly the same problem for the DS2000 when it's bandwidth is 300MHz; 200MHz is not an issue for it since it doesn't attempt to do sin(x)/x interpolation at sample rates below 1GSa/s (which puts the Nyquist frequency at 500MHz).

The major stipulation for sin(x)/x interpolation (which is a mathematical formula imposed on the sample points as opposed to just connecting the dots) to function correctly (i.e. not introduce errors into the reconstruction) is that NO components above the Nyquist frequency be present in the sampled signal. That is the MAIN rule that must be followed (to reap the benefit of a lower sample rate to frequency ratio) - there are reams of literature about it.

At 250MSa/s, the Nyquist frequency is 125MHz - and according to BW tests I've read about for the DS1000Z - frequencies >=125MHz do not seem to be adequately attenuated to make sin(x)/x interpolation a reliable option at that sample rate (i.e. there's no way to know for sure if it's introduced errors in the waveform).

So, just as the DS2000 automatically switches to linear interpolation at sample rates <=500MHz, I would suggest a DS1000Z owner that wants reliable signal reconstruction should switch to linear interpolation at sample rates <= 250MHz - or else - bandwidth-limit the signal to 20MHz. Either of those options, obviously, reduces the "effective reliably-interpolated maximum analog bandwidth" of the DSO when using all 4 channels.

Again, why?  If running at it's max 250MHz sample rate, that's certainly adequate for 100MHz on all 4 channels.  When sampling slower, of course, it's not.

You mean adequate when using an input filter designed to properly attenuate signals >=125MHz, don't you? Not only that, but a 2.5:1 ratio is actually fairly 'bottom-end':

From a LeCroy document on sin(x)/x interpolation:
"SinX interpolation works very well only when this ratio is greater than 2:1. 3:1 is a good ratio, with 4:1 usually working almost perfectly."

Again, the original 200MHz DS2000 only uses sin(x)/x when the ratio is 5:1 or greater.

I understood what you meant, though marmad did not.

I understood him perfectly fine, but I think he's wrong - at least in terms of the DS1000Z. He either doesn't understand how errors can be introduced into sin(x)/x reconstruction, hasn't thought about the filter roll-off of the DS1000Z, believes a sample rate 2.5x the highest frequency is good enough for linear interpolation, or is just speaking in non-specific, general terms. I'm guessing the latter, since that seems to be his MO on the forum in numerous threads.

In theory, 250MSa/s is (just) fast enough for a sin(x)/x reconstruction of 100MHz - but in the real (Rigol-designed-input-filter) world, I don't believe it is.

[Mark_O]

marmad, thanks for taking the time to clarify.

Actually, I've been suggesting exactly the same problem for the DS2000 when it's bandwidth is 300MHz;

Yes, I did see that.  Which most would apparently prefer to ignore.  "Hey!  I've got a 300MHz scope!"

200MHz is not an issue for it since it doesn't attempt to do sin(x)/x interpolation at sample rates below 1GSa/s...

Thanks.  I wasn't aware of that.  So it does linear reconstruction for 200MHz signals, when sampled at 500MSa/s. 

I wondered how that differed from the 1000Z though, also in linear mode for 100MHz signals, sampled at 250MSa/s?  You wrote, "...2.5 samples per period are not enough for an accurate linear reconstruction, pretty or not."

[This is one example of what puzzled me about your original posts.  You appeared to be singling out the 1000-series for criticisms, that applied equally well to the 2000; which you'd never been concerned about.  E.g., I had never heard you say about a DS2202, "it's not a 2-channel 200MHz BW DSO". 

But of course, the DS2000 goes one step further, by offering double the sample rate, because it's not trying to mux one ADC 4 ways (to save costs).  So there you get 5 samples per period, instead of 2.5, and that's enough for reliable sin(x)/x reconstruction of a 200MHz signal.  So now I understand what you were getting at.]

The major stipulation for sin(x)/x interpolation (which is a mathematical formula imposed on the sample points as opposed to just connecting the dots) to function correctly (i.e. not introduce errors into the reconstruction) is that NO components above the Nyquist frequency be present in the sampled signal. That is the MAIN rule that must be followed (to reap the benefit of a lower sample rate to frequency ratio) - there are reams of literature about it.

Yep.  Thanks for the review of Sampling Theory 101. 

At 250MSa/s, the Nyquist frequency is 125MHz - and according to BW tests I've read about for the DS1000Z - frequencies >=125MHz do not seem to be adequately attenuated to make sin(x)/x interpolation a reliable option at that sample rate (i.e. there's no way to know for sure if it's introduced errors in the waveform).

You may be right about the 1074Z, rated at 70MHz, but actually only -3dB ~90MHz.  You're certainly correct about the "100MHz" 1104Z, which is actually -3dB at ~160MHz.  250MSa/s is nowhere near adequate for sin(x)/x curve-fitting you can believe in, on a real or modded 1104Z.  And therefore shouldn't be used in 4-channel mode.

So, just as the DS2000 automatically switches to linear interpolation at sample rates <=500MHz, I would suggest a DS1000Z owner that wants reliable signal reconstruction should switch to linear interpolation at sample rates <= 250MHz - or else - bandwidth-limit the signal to 20MHz. Either of those options, obviously, reduces the "effective reliably-interpolated maximum analog bandwidth" of the DSO when using all 4 channels.

It would certainly be nice to have some better control over the BW Limiting, especially considering the realities of sampling constraints.  It's too bad the attenuator chip Rigol uses has only 20MHz and 100MHz rolloffs, at the low end.   80MHz and 50MHz would be handy to have. 

You mean adequate when using an input filter designed to properly attenuate signals >=125MHz, don't you? Not only that, but a 2.5:1 ratio is actually fairly 'bottom-end':

Yes.  And, yes.

From a LeCroy document on sin(x)/x interpolation:
"SinX interpolation works very well only when this ratio is greater than 2:1. 3:1 is a good ratio, with 4:1 usually working almost perfectly."

Again, the original 200MHz DS2000 only uses sin(x)/x when the ratio is 5:1 or greater.

Thanks!  I've read a lot of LeCroy's stuff, but don't recall this.  I liked the "usually", and "almost", with 4x.    I would have expected them to say, "works very well" with >2.5:1 though, rather than 2:1.  LeCroy uses good filters, but they're not brick-wall (because they'd introduce too much phase-shift in the pass-band).

In theory, 250MSa/s is (just) fast enough for a sin(x)/x reconstruction of 100MHz - but in the real (Rigol-designed-input-filter) world, I don't believe it is.

After reading your additional commentary, I'd have to agree with that conclusion.     Thanks again.

[marmad]

I wondered how that differed from the 1000Z though, also in linear mode for 100MHz signals, sampled at 250MSa/s?  You wrote, "...2.5 samples per period are not enough for an accurate linear reconstruction, pretty or not."

Well, whenever I'm using linear interpolation rates (<= 500MSa/s on the DS2000), I'm assuming that the effective reproducible bandwidth is 8-10x less than the sample rate. Fortunately, because of the memory depth of the DS2000, I can use the maximum sample rate (1-2GSa/s & sin(x)/x) with both channels on down to 2ms/div, if I so desire. But, as you mentioned, because of the single ADC, this is impossible on the DS1000Z with 3-4 channels on.